Implement Specialization Constants

* Added support to our local copy of SpirV Reflect (which does not support it).
* Pass them on render or compute pipeline creation.
* Not implemented in our shaders yet.

2 files changed, 164 insertions, 6 deletions

diff --git a/drivers/vulkan/rendering_device_vulkan.cpp b/drivers/vulkan/rendering_device_vulkan.cpp
index 6c1f1e4852..d3d49503d8 100644
--- a/drivers/vulkan/rendering_device_vulkan.cpp
+++ b/drivers/vulkan/rendering_device_vulkan.cpp
@@ -4374,6 +4374,8 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages
 
 	uint32_t stages_processed = 0;
 
+	Vector<Shader::SpecializationConstant> specialization_constants;
+
 	bool is_compute = false;
 
 	uint32_t compute_local_size[3] = { 0, 0, 0 };
@@ -4560,6 +4562,62 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages
 				}
 			}
 
+			{
+				//specialization constants
+
+				uint32_t sc_count = 0;
+				result = spvReflectEnumerateSpecializationConstants(&module, &sc_count, nullptr);
+				ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(),
+						"Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed enumerating specialization constants.");
+
+				if (sc_count) {
+					Vector<SpvReflectSpecializationConstant *> spec_constants;
+					spec_constants.resize(sc_count);
+
+					result = spvReflectEnumerateSpecializationConstants(&module, &sc_count, spec_constants.ptrw());
+					ERR_FAIL_COND_V_MSG(result != SPV_REFLECT_RESULT_SUCCESS, RID(),
+							"Reflection of SPIR-V shader stage '" + String(shader_stage_names[p_stages[i].shader_stage]) + "' failed obtaining specialization constants.");
+
+					for (uint32_t j = 0; j < sc_count; j++) {
+						int32_t existing = -1;
+						Shader::SpecializationConstant sconst;
+						sconst.constant.constant_id = spec_constants[j]->constant_id;
+						switch (spec_constants[j]->constant_type) {
+							case SPV_REFLECT_SPECIALIZATION_CONSTANT_BOOL: {
+								sconst.constant.type = PIPELINE_SPECIALIZATION_CONSTANT_TYPE_BOOL;
+								sconst.constant.bool_value = spec_constants[j]->default_value.int_bool_value != 0;
+							} break;
+							case SPV_REFLECT_SPECIALIZATION_CONSTANT_INT: {
+								sconst.constant.type = PIPELINE_SPECIALIZATION_CONSTANT_TYPE_INT;
+								sconst.constant.int_value = spec_constants[j]->default_value.int_bool_value;
+							} break;
+							case SPV_REFLECT_SPECIALIZATION_CONSTANT_FLOAT: {
+								sconst.constant.type = PIPELINE_SPECIALIZATION_CONSTANT_TYPE_FLOAT;
+								sconst.constant.float_value = spec_constants[j]->default_value.float_value;
+							} break;
+						}
+						sconst.stage_flags = 1 << p_stages[i].shader_stage;
+
+						print_line("spec constant " + itos(i) + ": " + String(spec_constants[j]->name) + " type " + itos(spec_constants[j]->constant_type) + " id " + itos(spec_constants[j]->constant_id));
+
+						for (int k = 0; k < specialization_constants.size(); k++) {
+							if (specialization_constants[k].constant.constant_id == sconst.constant.constant_id) {
+								ERR_FAIL_COND_V_MSG(specialization_constants[k].constant.type != sconst.constant.type, RID(), "More than one specialization constant used for id (" + itos(sconst.constant.constant_id) + "), but their types differ.");
+								ERR_FAIL_COND_V_MSG(specialization_constants[k].constant.int_value != sconst.constant.int_value, RID(), "More than one specialization constant used for id (" + itos(sconst.constant.constant_id) + "), but their default values differ.");
+								existing = k;
+								break;
+							}
+						}
+
+						if (existing > 0) {
+							specialization_constants.write[existing].stage_flags |= sconst.stage_flags;
+						} else {
+							specialization_constants.push_back(sconst);
+						}
+					}
+				}
+			}
+
 			if (stage == SHADER_STAGE_VERTEX) {
 				uint32_t iv_count = 0;
 				result = spvReflectEnumerateInputVariables(&module, &iv_count, nullptr);
@@ -4656,6 +4714,7 @@ RID RenderingDeviceVulkan::shader_create(const Vector<ShaderStageData> &p_stages
 	shader.compute_local_size[0] = compute_local_size[0];
 	shader.compute_local_size[1] = compute_local_size[1];
 	shader.compute_local_size[2] = compute_local_size[2];
+	shader.specialization_constants = specialization_constants;
 
 	String error_text;
 
@@ -5651,7 +5710,7 @@ Vector<uint8_t> RenderingDeviceVulkan::buffer_get_data(RID p_buffer) {
 /**** RENDER PIPELINE ****/
 /*************************/
 
-RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const PipelineRasterizationState &p_rasterization_state, const PipelineMultisampleState &p_multisample_state, const PipelineDepthStencilState &p_depth_stencil_state, const PipelineColorBlendState &p_blend_state, int p_dynamic_state_flags, uint32_t p_for_render_pass) {
+RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const PipelineRasterizationState &p_rasterization_state, const PipelineMultisampleState &p_multisample_state, const PipelineDepthStencilState &p_depth_stencil_state, const PipelineColorBlendState &p_blend_state, int p_dynamic_state_flags, uint32_t p_for_render_pass, const Vector<PipelineSpecializationConstant> &p_specialization_constants) {
 	_THREAD_SAFE_METHOD_
 
 	//needs a shader
@@ -5969,8 +6028,63 @@ RID RenderingDeviceVulkan::render_pipeline_create(RID p_shader, FramebufferForma
 	graphics_pipeline_create_info.pNext = nullptr;
 	graphics_pipeline_create_info.flags = 0;
 
-	graphics_pipeline_create_info.stageCount = shader->pipeline_stages.size();
-	graphics_pipeline_create_info.pStages = shader->pipeline_stages.ptr();
+	Vector<VkPipelineShaderStageCreateInfo> pipeline_stages = shader->pipeline_stages;
+	Vector<VkSpecializationInfo> specialization_info;
+	Vector<Vector<VkSpecializationMapEntry>> specialization_map_entries;
+	Vector<uint32_t> specialization_constant_data;
+
+	if (shader->specialization_constants.size()) {
+		specialization_constant_data.resize(shader->specialization_constants.size());
+		uint32_t *data_ptr = specialization_constant_data.ptrw();
+		specialization_info.resize(pipeline_stages.size());
+		specialization_map_entries.resize(pipeline_stages.size());
+		for (int i = 0; i < shader->specialization_constants.size(); i++) {
+			//see if overriden
+			const Shader::SpecializationConstant &sc = shader->specialization_constants[i];
+			data_ptr[i] = sc.constant.int_value; //just copy the 32 bits
+
+			for (int j = 0; j < p_specialization_constants.size(); j++) {
+				const PipelineSpecializationConstant &psc = p_specialization_constants[j];
+				if (psc.constant_id == sc.constant.constant_id) {
+					ERR_FAIL_COND_V_MSG(psc.type != sc.constant.type, RID(), "Specialization constant provided for id (" + itos(sc.constant.constant_id) + ") is of the wrong type.");
+					data_ptr[i] = sc.constant.int_value;
+					break;
+				}
+			}
+
+			VkSpecializationMapEntry entry;
+
+			entry.constantID = sc.constant.constant_id;
+			entry.offset = i * sizeof(uint32_t);
+			entry.size = sizeof(uint32_t);
+
+			for (int j = 0; j < SHADER_STAGE_MAX; j++) {
+				if (sc.stage_flags & (1 << j)) {
+					VkShaderStageFlagBits stage = shader_stage_masks[j];
+					for (int k = 0; k < pipeline_stages.size(); k++) {
+						if (pipeline_stages[k].stage == stage) {
+							specialization_map_entries.write[k].push_back(entry);
+						}
+					}
+				}
+			}
+		}
+
+		for (int k = 0; k < pipeline_stages.size(); k++) {
+			if (specialization_map_entries[k].size()) {
+				specialization_info.write[k].dataSize = specialization_constant_data.size() * sizeof(uint32_t);
+				specialization_info.write[k].pData = data_ptr;
+				specialization_info.write[k].mapEntryCount = specialization_map_entries[k].size();
+				specialization_info.write[k].pMapEntries = specialization_map_entries[k].ptr();
+
+				pipeline_stages.write[k].pSpecializationInfo = specialization_info.ptr();
+			}
+		}
+	}
+
+	graphics_pipeline_create_info.stageCount = pipeline_stages.size();
+	graphics_pipeline_create_info.pStages = pipeline_stages.ptr();
+
 	graphics_pipeline_create_info.pVertexInputState = &pipeline_vertex_input_state_create_info;
 	graphics_pipeline_create_info.pInputAssemblyState = &input_assembly_create_info;
 	graphics_pipeline_create_info.pTessellationState = &tessellation_create_info;
@@ -6039,7 +6153,7 @@ bool RenderingDeviceVulkan::render_pipeline_is_valid(RID p_pipeline) {
 /**** COMPUTE PIPELINE ****/
 /**************************/
 
-RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader) {
+RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader, const Vector<PipelineSpecializationConstant> &p_specialization_constants) {
 	_THREAD_SAFE_METHOD_
 
 	//needs a shader
@@ -6061,6 +6175,44 @@ RID RenderingDeviceVulkan::compute_pipeline_create(RID p_shader) {
 	compute_pipeline_create_info.basePipelineHandle = VK_NULL_HANDLE;
 	compute_pipeline_create_info.basePipelineIndex = 0;
 
+	VkSpecializationInfo specialization_info;
+	Vector<VkSpecializationMapEntry> specialization_map_entries;
+	Vector<uint32_t> specialization_constant_data;
+
+	if (shader->specialization_constants.size()) {
+		specialization_constant_data.resize(shader->specialization_constants.size());
+		uint32_t *data_ptr = specialization_constant_data.ptrw();
+		for (int i = 0; i < shader->specialization_constants.size(); i++) {
+			//see if overriden
+			const Shader::SpecializationConstant &sc = shader->specialization_constants[i];
+			data_ptr[i] = sc.constant.int_value; //just copy the 32 bits
+
+			for (int j = 0; j < p_specialization_constants.size(); j++) {
+				const PipelineSpecializationConstant &psc = p_specialization_constants[j];
+				if (psc.constant_id == sc.constant.constant_id) {
+					ERR_FAIL_COND_V_MSG(psc.type != sc.constant.type, RID(), "Specialization constant provided for id (" + itos(sc.constant.constant_id) + ") is of the wrong type.");
+					data_ptr[i] = sc.constant.int_value;
+					break;
+				}
+			}
+
+			VkSpecializationMapEntry entry;
+
+			entry.constantID = sc.constant.constant_id;
+			entry.offset = i * sizeof(uint32_t);
+			entry.size = sizeof(uint32_t);
+
+			specialization_map_entries.push_back(entry);
+		}
+
+		specialization_info.dataSize = specialization_constant_data.size() * sizeof(uint32_t);
+		specialization_info.pData = data_ptr;
+		specialization_info.mapEntryCount = specialization_map_entries.size();
+		specialization_info.pMapEntries = specialization_map_entries.ptr();
+
+		compute_pipeline_create_info.stage.pSpecializationInfo = &specialization_info;
+	}
+
 	ComputePipeline pipeline;
 	VkResult err = vkCreateComputePipelines(device, VK_NULL_HANDLE, 1, &compute_pipeline_create_info, nullptr, &pipeline.pipeline);
 	ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateComputePipelines failed with error " + itos(err) + ".");
diff --git a/drivers/vulkan/rendering_device_vulkan.h b/drivers/vulkan/rendering_device_vulkan.h
index ff9ad71268..8b95ff43b8 100644
--- a/drivers/vulkan/rendering_device_vulkan.h
+++ b/drivers/vulkan/rendering_device_vulkan.h
@@ -623,11 +623,17 @@ class RenderingDeviceVulkan : public RenderingDevice {
 
 		uint32_t compute_local_size[3] = { 0, 0, 0 };
 
+		struct SpecializationConstant {
+			PipelineSpecializationConstant constant;
+			uint32_t stage_flags = 0;
+		};
+
 		bool is_compute = false;
 		int max_output = 0;
 		Vector<Set> sets;
 		Vector<uint32_t> set_formats;
 		Vector<VkPipelineShaderStageCreateInfo> pipeline_stages;
+		Vector<SpecializationConstant> specialization_constants;
 		VkPipelineLayout pipeline_layout = VK_NULL_HANDLE;
 	};
 
@@ -1100,14 +1106,14 @@ public:
 	/**** RENDER PIPELINE ****/
 	/*************************/
 
-	virtual RID render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const PipelineRasterizationState &p_rasterization_state, const PipelineMultisampleState &p_multisample_state, const PipelineDepthStencilState &p_depth_stencil_state, const PipelineColorBlendState &p_blend_state, int p_dynamic_state_flags = 0, uint32_t p_for_render_pass = 0);
+	virtual RID render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const PipelineRasterizationState &p_rasterization_state, const PipelineMultisampleState &p_multisample_state, const PipelineDepthStencilState &p_depth_stencil_state, const PipelineColorBlendState &p_blend_state, int p_dynamic_state_flags = 0, uint32_t p_for_render_pass = 0, const Vector<PipelineSpecializationConstant> &p_specialization_constants = Vector<PipelineSpecializationConstant>());
 	virtual bool render_pipeline_is_valid(RID p_pipeline);
 
 	/**************************/
 	/**** COMPUTE PIPELINE ****/
 	/**************************/
 
-	virtual RID compute_pipeline_create(RID p_shader);
+	virtual RID compute_pipeline_create(RID p_shader, const Vector<PipelineSpecializationConstant> &p_specialization_constants = Vector<PipelineSpecializationConstant>());
 	virtual bool compute_pipeline_is_valid(RID p_pipeline);
 
 	/****************/